Automating notification creation with LLM
Designing an automated notification creation flow that produces high-quality, compliant messages at scale, aligned with messaging and policy requirements.
Platform:
Meta Horizon
Scope:
UX writing, prompt design, policy aware.
Partners:
Engineering, product design, policy, and legal stakeholders
The problem
As part of building successful games, creators use engagement mechanics such as challenges, events, and leaderboards to bring players back into the experience. Notifications are a key part of this loop, prompting players to return and take action.
However, creating these messages manually required review from policy and legal teams to ensure compliance, in addition to meeting content quality standards. This made the process slow and inconsistent — approval cycles could take up to two working days — and messages sometimes failed to meet both quality and regulatory requirements on first submission.
The process
Cross-functional definition of the notification flow
I partnered with engineering and product design to define an automated notification creation flow that evaluates engagement mechanisms and generates AI-written messages within a strict content framework.
Building compliant AI generation with guardrails
Working with engineering, I helped define a synchronous LLM system embedding content criteria, policy requirements, and a dos-and-don’ts framework. This enabled instant generation of multiple pre-approved notification options, reducing reliance on manual review cycles.
Designing creator control and iteration pathways
I designed editing capabilities allowing creators to refine AI-generated notifications before sending, with edits triggering review to maintain compliance. I also prototyped the end-to-end flow using AI tools to rapidly test and iterate.
The result
Faster notification creation at scale
Reduced time to generate compliant notifications from hours to seconds, with creators using AI-generated message that met policy guidelines, instead of writing them from scratch.
Maintained compliance without slowing production
Embedded guardrails ensured messages met policy and quality standards while reducing reliance on repeated manual review cycles.
Clearer approval model for creators
Introduced a transparent flow that separated pre-approved AI outputs from edited content, making compliance requirements easier to understand and follow.